One of a major challenge in IDS is to discover thernintrusive patterns which are normally hidden in abundantrnof data. Furthermore, it has many features. Some of thernfeatures are redundant and some are less significant andrnthey contribute little to the detection process. The purposernof this study is to identify an optimum number ofrnsignificant features that can represent each category;rnNormal, Probe, U2R, R2L and DoS. Here, we deployedrnhierarchical feature selection approach and usedrnsimilarity-based classification (Kohonen Self-OrganizingrnMap) to classify an input data into their respectiverncategories. Performance was measured based on theirrncorrect classification. Empirical results suggest that therernis no generic feature subset which is suitable to representrnall categories. Instead, different categories are bestrnrepresented using different feature subsets.
展开▼